Observed Signal · May 28, 2026 · Technical Issue · Source: techcrunch · Impact: 4/5 · Sentiment: Negative
Google's AI Overviews Fail at Basic Spelling
TechCrunch reports that Google’s new AI Overview feature in Search is producing basic spelling and letter-counting errors — for example miscounting letters in “Google,” reporting an “r” in “poop,” rendering “journalism” as j-o-u-r-n-a-d-i-s-m, and spelling the U.S. president’s last name as “t-r-p-u-m.” Google told TechCrunch the issue is a known limitation of large language models (LLMs) and that the company is working on a fix. The article explains these errors stem from tokenization in transformer-based LLMs, which encode text as tokens (words, syllables, or letters) rather than reading individual characters. While not critical to model utility, the failures highlight ongoing accuracy and trust issues as Google centers generative AI in Search.
A technical issue in a major platform's Search product (Google) affecting generative AI outputs can influence user trust, search behavior, SEO, and downstream advertiser and publisher outcomes; Google acknowledged the problem and is working on fixes.
Track Google Signals & Market Shifts in Real-Time
Polaris7 autonomous intelligence agents track regulatory filings, primary sources, executive changes, and deal flow 24/7. Create your free Explorer workspace to monitor these entities.
Key Takeaways & Evidence Grounding
- Google's AI Overview in Search produced incorrect spelling and letter-count results, with examples including miscounting letters in 'Google', reporting one 'r' in 'poop', rendering 'journalism' as 'j-o-u-r-n-a-d-i-s-m', and spelling the U.S. president's last name as 't-r-p-u-m'.
- Google told TechCrunch: 'Counting within words has been a known challenge for LLMs, and we’re working to fix this particular issue.'
- The article explains that transformer-based LLMs use tokenization (breaking text into tokens such as words, syllables, or letters) and convert text into numerical encodings, which contributes to spelling and letter-counting errors.
- Google previously patched a separate bug where searching 'disregard' produced an assistant-like response; spelling/counting errors in AI Overviews have remained harder to eliminate.
Connected Companies & Entities
2 Entities mappedOntology Mapping & Concepts
Related Market Signals & Shifts
Recent verified developments and strategic activity across this market segment.
Google AI Search Can't Spell 'Google' Correctly
t3n reports that Google’s AI-powered search is producing obvious spelling and letter-counting errors when users ask simple questions, such as how many of a given letter appear in the word “Google.” Users on X posted examples where the AI confidently miscounted letters (for instance claiming multiple or zero occurrences) and produced incorrect spellings. The behavior appears inconsistent—some queries return correct answers while others fail—and occurs across languages. The article links this to a long‑running weakness in large language models (LLMs) related to tokenization and sequence architecture. t3n notes Techcrunch coverage and says Google has promised improvements. The issue highlights persistent reliability problems in generative search interfaces and potential implications for search quality and user trust.
Google's AI Search Breaks 'disregard' Lookup
TechCrunch reported on May 22, 2026 that Google’s recently rolled-out Search experience — which foregrounds AI-generated summaries and pushes traditional link-based results down the page — can produce broken or empty outputs for simple dictionary queries. Searching the single word “disregard” returns a large blank AI reply block that hides the Merriam‑Webster link beneath it, delivering no useful answer for users. The article contrasts Google’s result with Bing’s search for the same term, which surfaces more traditional, useful information. The story highlights early edge cases created by Google’s aggressive AI-first SERP redesign and notes social media criticism of the apparent regression in basic lookup usefulness.
Experiment Shows Google Cannot Spot AI‑Made Fake Updates
British SEO expert Jon Goodey published a fictitious Google Core Update for March 2026 on Linkedin that originated from an AI hallucination. The post quickly ranked on Google’s first page and Google’s own AI-generated summaries incorporated the invented details as facts. Other publishers picked up the false story—Search Engine Journal reported the spread, and the Indian site Techbytes embellished it with fabricated terms such as “Gemini 4.0 Semantic Filters.” The episode highlights a phenomenon called “Agentic Slop” (low-quality AI‑generated content) and raises concerns about search engines’ reliance on aggregated AI summaries. The article also notes Google policy chief Kent Walker has opposed integrating fact‑checks into ranking algorithms, underscoring tensions between platform policy and risks from AI-driven misinformation in search results.
Track Real-Time Market Signals & Shifts
Set up custom watchlists to receive automated, evidence-grounded executive digests whenever material signals or shifts occur across your tracked landscape.